Baseline Statistics Outputting Same FDR
Documentation and individual testing revealed FDR is the same despite different p-values; graphs seem to support that FDR's should be different. Attached is object and code to reproduce:
shazam::testBaseline(grouped_1, groupBy = "foxp3")
[thanks to Yisi for the find]
Comments (5)
-
-
reporter Okay, I trust in the stats (esp since I haven't run through the formulas myself), just seems fishy to that the FDR values are the same across both regions, so looked like something is wrong in the calculation..
-
Yeah.
Though, we should check that it's not a tibble indexing issue, where we ending up with only the first value instead of a vector of values being assigned into a column.
-
Adjusting for multiple testing is done via p.adjust() with method="fdr", which is an alias for "BH". Numerically identical q-values are not erroneous due to the nature of the Benjamin-Hochberg procedure.
See link before for more details.
https://stats.stackexchange.com/questions/263831/over-represented-values-in-fdr-adjusted-p-values
-
reporter - changed status to resolved
Just tested with a new dataset - not a tibble copying error.
- Log in to comment
This is unlikely a bug but rather how FDR is expected to behave.